Not waiting forever.
WHY TIMEOUTS MATTER MORE THAN THEY SEEM
A call with no timeout can hold a connection, a thread and a transaction indefinitely.
WHAT THAT CAUSES
Resource exhaustion in your own system, caused by someone else's.
WHAT TO SET
A connection timeout A read timeout A total time limit
WHY SEPARATE
Connecting quickly then hanging is a common failure.
WHAT VALUES TO CHOOSE
Based on observed response times, not guesses.
HOW TO ESTABLISH THEM
Measure the distribution, and set above the high percentile.
WHY NOT THE AVERAGE
Most calls are fast, and the timeout is about the slow ones.
WHAT TO CONSIDER ABOUT NESTED CALLS
Each layer's timeout must be shorter than the layer above.
WHY
Otherwise the outer layer gives up while the inner still waits.
WHAT THAT PATTERN IS CALLED
A timeout budget.
WHAT TO DO ON TIMEOUT
Treat the outcome as unknown, not as failure.
WHY THAT DISTINCTION IS CRITICAL
The other side may have completed the work.
WHAT THAT MEANS FOR PAYMENTS
Never assume a timed-out payment did not happen.
WHAT TO DO INSTEAD
Query its status, or reconcile.
WHAT TO LOG
Timeouts separately from errors.
WHY
They indicate different problems and different responses.
